I'm analyzing the cloud image, the code I'm using is linked here go.pro
I apologize for the dark background and terrible color scheme. (I'm sure I loaded black and white color table but everything came out red...Let's just pretend it's a Halloween night...) try this to cure the red problem: device, decompose=0
First, fft the image. The wave number of this picture are concentrated in small values, both on x and y axises.
Filter out all other wave numbers except the ones between 10 and 100. Inverse the filtered power spectrum, we see a somewhat compromised image as below (right). Neither large scale nor small scale patterns are clear.
Sum the PSD in each total wave number bin (of size 1), I have the relationship between total wavenumber and the variance.
It seems the -5/3 law still can apply in this case, but the -3 doesn't fit quite well.
Now let's play with the butterfly image. (love butterflies!)
This time, apply a low-pass filter, and the inverse image shows only the large scale (small wave number) pattern.
I apologize for the dark background and terrible color scheme. (I'm sure I loaded black and white color table but everything came out red...Let's just pretend it's a Halloween night...)
try this to cure the red problem: device, decompose=0
First, fft the image. The wave number of this picture are concentrated in small values, both on x and y axises.
Filter out all other wave numbers except the ones between 10 and 100. Inverse the filtered power spectrum, we see a somewhat compromised image as below (right). Neither large scale nor small scale patterns are clear.
Sum the PSD in each total wave number bin (of size 1), I have the relationship between total wavenumber and the variance.
It seems the -5/3 law still can apply in this case, but the -3 doesn't fit quite well.
Now let's play with the butterfly image. (love butterflies!)
This time, apply a low-pass filter, and the inverse image shows only the large scale (small wave number) pattern.